Neural Network Architectures for Location Estimation in the Internet of Things

TL;DR

This paper introduces a novel AI-based wireless location estimation method that leverages the Cramer-Rao bound to optimize neural network design, resulting in efficient, accurate, and less overfit solutions applicable to IoT and vehicular networks.

Contribution

It is the first to integrate the Cramer-Rao bound into neural network design for wireless localization, improving efficiency and accuracy in AI solutions.

Findings

Neural network size can be optimized using the Cramer-Rao bound.

Proposed algorithms outperform traditional methods in localization accuracy.

Approach reduces overfitting and computational time.

Abstract

Artificial Intelligence (AI) solutions for wireless location estimation are likely to prevail in many real-world scenarios. In this work, we demonstrate for the first time how the Cramer-Rao upper bound on localization accuracy can facilitate efficient neural-network solutions for wireless location estimation. In particular, we demonstrate how the number of neurons for the network can be intelligently chosen, leading to AI location solutions that are not time-consuming to run and less likely to be plagued by over-fitting. Experimental verification of our approach is provided. Our new algorithms are directly applicable to location estimates in many scenarios including the Internet of Things, and vehicular networks where vehicular GPS coordinates are unreliable or need verifying. Our work represents the first successful AI solution for a communication problem whose neural-network design is based on fundamental information-theoretic constructs. We anticipate our approach will be useful for a wide range of communication problems beyond location estimation.